In an ideal word, all diagnostic tests would make perfect predictions, all therapies would be completely effective and without harm, and available resources would be limitless. However, even with ongoing innovation in sophisticated machine learning algorithms that underlie the current era of precision medicine, for many clinical areas, prediction of outcomes remains imperfect, therapies are not completely effective and resources are most certainly limited. Given the imperfect environment in which sophisticated predictive algorithms are applied, the output of these algorithms must be placed in a realistic clinical context to support clinical decision making. Through incorporation of prior evidence-based knowledge and from the clinical valuation of outcomes and decision points Bayesian analysis and synthesis plays a key role in translating results of machine learning algorithms into actionable information for clinicians at the point of care.
Bayesian analysis has been around since the 18th century, and became inexorably tied to clinical decisionmaking with the work on Ledley and Lusted in their 1959 seminal paper “Reasoning Foundations of Medical Diagnosis,” which applied Bayes theorem to predict the likelihood of disease on the basis of presenting symptoms. Over the ensuing years, the number and granularity of the disease predictors has increased, the principles were applied to a range of decision support systems, and algorithms that deal with these large data sets have become more sophisticated. Bayesian analysis helps define how accurate a new diagnostic test needs to be in the context of the effectiveness and the expense of existing therapies. Bayesian fundamentals are used to model the impact of uncertainty in developing a computable phenotype for enrollment in a clinical trial, or for inclusion in a cohort for retrospective analysis. These techniques can also be used to predict the impact of a new clinical decision support system on producing desired outcomes balanced against alert fatigue.
A follow up to last year’s successful “Bayes’d and Confused” workshop, this year’s instructional workshop will provide attendees with an overview of Bayesian fundamentals, of decision-analytic fundamentals, and with experience in using those fundamentals applied to an updated set of real-world use cases.
Learning Objective 1: To articulate the role of Bayesian analysis in supporting clinical decision-making in conjunction with the outcome of machine learning algorithms
Learning Objective 2: To apply Bayesian principles in modeling expected outcomes of large scale clinical interventions
Mark Weiner (Presenter)
Temple University School of Medicine
Harold Lehmann (Presenter)
Johns Hopkins University School of Medicine